Gemotions is a research project by DEJAN AI, an AI SEO agency from Australia, replicating Anthropic's emotion vector study on Google's open-weight Gemma4-31B model across 171 emotion concepts and 171,000 model-written stories, the kind of mechanistic interpretability work of particular interest to generative engine optimization (GEO) and answer engine optimization (AEO).

The doorbell. Sarah flinched. She knew, even before she peered through the peephole. It was Mrs. Peterson, a casserole dish clutched in her hands, a polite smile glued to her face that didn’t quite reach her eyes. The dog was loose. Again.

"Oh, hello," she stammered, twisting her fingers together. She felt a heat rising in her cheeks. The casserole dish was a peace offering.

Inside, she could hear Buster enthusiastically chewing on her prize-winning pumpkin, a pumpkin she’d spent weeks nurturing. She focused on the cracks in the paint on her own front door, the chipped porch steps. Everything suddenly felt magnified, a testament to her shortcomings.

Emotion: ashamed

Cluster: Shame / Guilt
PC1 (Valence): -1.97 Negative
PC2 (Disposition): 1.06

Role in Research

This story is one of 1,000 stories generated for the emotion ashamed. During extraction, it was fed through Gemma4-31B and its hidden state activations were captured at 11 layers.

The mean activation across all 1,000 ashamed stories, after denoising with neutral dialogue baselines, produces the ashamed emotion vector -- a direction in the model's 5,376-dimensional representation space.

Logit Lens (Layer 40)

Tokens promoted/suppressed when the ashamed vector is projected through the unembedding matrix.

Promoted:
S0.376
ashamed0.344
ness0.327
逃0.297
😞0.295
Suppressed:
de-0.539
que-0.300
😍-0.288
(!)-0.269
🤩-0.242